Reproduce “ResMAS: Resilience Optimization in LLM-based Multi-agent Systems” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Zhilun Zhou, Zihan Liu, Jiahe Liu, Qingyu Shao, Yihan Wang, Kun Shao, Depeng Jin, Fengli Xu

Researchers from Tsinghua University and Huawei Noah’s Ark Lab developed ResMAS, a two-stage framework that optimizes communication topology and agent prompts to enhance the resilience of LLM-based multi-agent systems against random agent failures. This approach consistently demonstrates higher resilience scores compared to baselines across multiple tasks and generalizes effectively to unseen scenarios and different LLM backbones.

Key insights

  • LLM-based Multi-Agent Systems exhibit inherent resilience to perturbations, which generally increases with the number of agents and communication links.
  • Both the communication topology and the design of individual agent prompts profoundly influence an MAS's ability to withstand failures, even when the number of agents and links are held constant.
  • Proactively designing MAS structures for inherent resilience against common random failures is critical for dependable deployment, moving beyond reactive defense mechanisms.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.